Wrap Up

1
Wrap Up

• Introduction toArtificial Intelligence
• COS302
• Michael L. Littman
• Fall 2001

2
Administration

• Laptop four is busted.
• Course evaluations.

3
Plan

• LSI and autoassociation
• Topic list
• Second half Themes
• Wrap up

4
Passage Autoassociator

• word features
• capture gist so it can be reconstructed

5
Well Make Computers Brighter (sorry Billy Joel)

• Gang and Kedar relaxation color graphs by
  propagation
  
• Neighbor goal state Davis Putnam change to CNF
• CSP satisfaction forward checking giving
  traction
  
• BFS and DFS and A is the best

6
Verse 2

• Local search save time optimize hill climb
• Nim Sharks GSAT chess evaluation hacks
• Alpha beta game tree prune the branch and
  youll see
  
• Fitness function reproduction phase transition
  min-max

7
Chorus

• Well make computers brighter
• With their and and oring
• Theyll beat Alan Turing.
• Well make computers brighter
• Smarter than a porpoise
• Train it with a corpus

8
Verse 3

• Markov model Chinese Room Nim-Rand in a matrix
  form
  
• Labeled data, missing data logic of Boole
• Discount rank list Altavista Wordnet
• IR bag-of-words flip around with Bayes Rule

9
Verse 4

• Rap song n-queen Susan swallowed something
  green
  
• Language model classify take a max and
  multiply
  
• Zipfs law trigrams smooth it out with
  bigrams
  
• Naïve Bayes winning plays write another program

10
Chorus

• Well make computers brighter
• With their and and oring
• Theyll beat Alan Turing.
• Well make computers brighter
• Smarter than a porpoise
• Train it with a corpus

11
Verse 5

• HMM part-of-speech they will learn if you
  will teach
  
• Baum-Welch if you please Rush Hour and
  Analogies
  
• Weighted coins network lags Viterbi finds most
  likely tags
  
• Start running watch for ties expectation
  maximize

12
Verse 6

• Learning function feature try learning whos a
  girl
  
• overfit PAC bounds build your own decision
  tree
  
• Too big memorize cross validate to
  generalize
  
• supervising pure leaf fit the function to a t

13
Chorus

• Well make computers brighter
• With their and and oring
• Theyll beat Alan Turing.
• Well make computers brighter
• Smarter than a porpoise
• Train it with a corpus

14
Verse 7

• Delta rule backprop hidden unit dont stop
• Learn pairs sum squares linear perceptron
• Error surface is quite bent just descend the
  gradient and-or sigmoid nonlinear regression

15
Bridge/Chorus

• XOR learning rate squash the sum and
  integrate
  
• Neural net split the set How much better can
  it get?
  
• Chorus Well make computers brighter
• With their and and oring
• Theyll beat Alan Turing.
• Well make computers brighter
• Smarter than a porpoise
• Train it with a corpus

16
Verse 8

• LSI max and min local search is back again
• TOEFL pick best essay grade pass the test
• Matrix Plato SVD vectors all in high D
• All the points are in the space dont you make an
  Eigenface!

17
Verse 9

• Segmentation Saturn spin Probabilities a win
• Q and P iteratively Bayes net CPT
• Autopilots in the sky
• Play chess masters for a tie
• Alan Alda robots cry
• All of this is in AI!

18
Chorus

• Well make computers brighter
• With their and and oring
• Theyll beat Alan Turing.
• Well make computers brighter
• Smarter than a porpoise
• Train it with a corpus

19
Final Chorus

• Well make computers brighter
• Hope you had some fun
• Because the class is done, is done, is done

20
Probability Models

• Compare and contrast
• unigram
• bigram
• trigram
• HMM
• belief net
• Naïve Bayes

21
Classification Algs

• Compare and contrast
• Naïve Bayes
• decision tree
• perceptron
• multilayer neural net

22
Missing Data

• Compare and contrast
• EM
• SVD
• gradient descent

23
Dynamic Programming

• Compare and contrast
• Viterbi
• forward-backward
• minimax
• Markov chains (value iteration)
• segmentation

24
Big Picture This is AI?

• We learned a bunch of programming tricks people
  still do the hard work.
  
• Yeah, thats true.
• Its not such a bad thing building more flexible
  software (search, learning).
  
• Were working towards more independent artifacts
  not there yet

25
Thank You!

• Thanks for your help!
• Gang and Kedar
• Lisa and the kids
• ATT
• Andrew Moore
• Family Feud survey respondents
• Peter Stone
• all of you!